Gen AI fabrics and the data architecture challenge facing AI – News

Generative artificial intelligence is no longer an experiment tucked away in innovation labs. Enterprises now push toward production-scale systems that can reason, retrieve and act across vast data estates. The shift exposes the fact that existing data architectures must evolve to meet the demands of today’s AI imperative. Gen AI fabrics are emerging as the connective layer enterprises now need to implement. Fragmented pipelines, brittle governance and cloud-first assumptions now intersect with workloads that demand low latency, locality and trust across environments.

What holds many organizations back is the structural gap between where enterprise data lives and how AI systems need to access it. Modern AI systems assume enterprise data can be discovered, governed and served consistently, wherever it resides. For many enterprises, that assumption doesn’t hold, according to Dave Vellante, chief analyst at theCUBE Research.

“AI [return on investment] isn’t being held back by model capability; rather, the friction point is data,” he said. “Our research shows that successful enterprise AI projects start with a foundational data infrastructure, ideally with open, composable fabrics, that can unify access, governance and resiliency across clouds, on-prem and out to the edge. The key, in our view, is to put the underlying data substrate in place and get on the AI curve as soon as possible. Allow AI to do the dirty work, learn from early projects and create a flywheel effect for subsequent use cases.”

This market reality underscores why gen AI fabrics are emerging as a distinct architectural layer. Rather than a single platform or product, these fabrics represent a shift toward open, composable data foundations designed to support AI at scale. Dell Technologies Inc.’s expanded AI Data Platform offers a clear example of how vendors are responding to this shift as evidence of where enterprise AI architecture is heading.

This feature is part of News Media’s exploration of how gen AI fabrics are reshaping enterprise data architecture for production-scale AI. (* Disclosure below.)

Data architecture: The real bottleneck from AI pilots to production

Enterprises want to move from AI experimentation to operational excellence. However, many falter when it comes time to scale. As generative and agentic workloads move closer to the core of the business, the limiting factor isn’t model access, but whether existing data environments can support production-grade AI, according to Varun Chhabra, senior vice president of ISG product marketing at Dell Technologies.

“Organizations are no longer just experimenting with AI; they’re increasingly deploying it to drive real business impact,” he said. “As they do that, we hear the same challenge again and again. The success of AI projects is often limited by the ability to unlock the value of enterprise data. Enterprise data has grown to a massive size. It’s increasingly fragmented and distributed and often locked in silos across workloads and locations.”

This production friction reflects a broader shift in how generative and agentic AI workloads pressure enterprise data infrastructure, according to Rob Strechay, principal analyst at theCUBE Research. As AI systems move closer to operational use, existing data environments are increasingly strained.

“Data infrastructure is once again under pressure from a new application,” he said. “This time, it is gen AI and agentic AI workflows requiring data be everywhere at all times. In our Future of Data Platforms Summit research, we find that scaling AI is a primary challenge for 65.4% of respondents and that 67% of data is stored in cloud or hybrid environments, with 33% still on-premises-only.”

Gen AI and agentic systems amplify long-standing data challenges rather than replacing them. As AI systems move closer to operational decision-making, the cost of inaccessible or poorly governed data rises sharply, according to both Vellante and Strechay.

“If you don’t have quality data that is accessible, you’re in a tough spot,” Strechay said. “A lot of people have struggled with fragmented, siloed data across multiple different systems, and AI really requires and operates across cloud, core and edge. I think what people are trying to do is [figure out], ‘How do we … bring the right tooling together?’”

How gen AI fabrics differ from traditional data platforms

As enterprises push AI closer to production systems, many are discovering that traditional data platforms struggle with the demands of generative and agentic workloads. Gen AI fabrics reflect a different architectural assumption in how data should be accessed and activated for AI, according to Travis Vigil, SVP of ISG product management at Dell Technologies.

“AI isn’t just about compute; it’s about the data that feeds it,” he said. “Most vendors would say push everything into a single storage layer, but this … slows things down and locks you in. The real challenge isn’t just storing the data; it’s activating it [and] turning raw information into something that’s usable, trustworthy and scalable for AI.”

As generative and agentic AI workloads evolve, enterprises are reconsidering how tightly coupled their data platforms need to be. Flexibility becomes as crucial as performance as AI pipelines grow more complex, according to Anahad Dhillon, director of product management at Dell.

“With our Dell AI Data Platform, storage engines and data engines are each optimized for what they do best, working together but independently,” he said. “That means no bottlenecks, no lock-in, just the freedom to innovate on your terms. By separating data processing from storage, we eliminate friction and reduce solution costs.”

This separation distinguishes gen AI fabrics from earlier generations of data platforms. Instead of assuming a fixed data gravity point, fabrics support more modular AI pipelines that can evolve over time. The metaphor of a fabric matters because it implies connection without consolidation, according to Vellante and Strechay.

“They’re not trying to silo you all into one box,” Strechay said. “The architecture lets customers plug in different pieces based on best of breed. They’re … letting that optionality be easy … based on your needs so that it can evolve over time.”

Dell, hyperscalers and the battle over AI data control

The rise of gen AI fabrics reflects a broader shift in how enterprises view control over AI infrastructure. After years of cloud-first strategies, many organizations are reexamining the trade-offs of moving all AI data and workloads into hyperscaler environments, according to Kevin Deierling, SVP of networking at Nvidia Corp.

“Unlocking AI’s full potential requires the right combination of compute, networking, software and storage,” he said. “Together [with Dell], we’ve delivered a unified Dell AI Data Platform with Nvidia, bringing Nvidia’s accelerated compute together with Dell’s composable storage and data engines. This makes AI faster to deploy, easier to scale and more cost-efficient.”

For many enterprises, gen AI fabrics offer a way to retain architectural choice while still supporting production-scale AI across environments. Rather than offering a vertically integrated AI stack, Dell positions itself as an infrastructure integrator, assembling validated components that can operate across cloud, core and edge. That approach resonates with organizations that want AI capabilities without surrendering control over where data lives and how it’s governed, according to Vellante and Strechay.

“Organizations put so much emphasis on the cloud that they … de-emphasized … their on-prem environments,” Vellante said. “Now they’re saying … ‘I want to bring that intelligence to the data that lives on-prem.’ I think they’re realizing that their data stack maybe needs to catch up to the modern data stacks that have been in the cloud … but to achieve that vision, they have to get their data in a place where they can serve it up in a governed manner to agents.”

As AI moves beyond inference into action, the demands placed on enterprise data infrastructure increase sharply. That shift is pushing organizations to rethink what production-ready AI infrastructure really looks like, according to Vellante.

“I’m envisioning the cloud as this expanding universe and AI overlaid around … but then agents being able to take advantage of that governed data and actually take actions with humans in the loop,” he said. “I think AI factories are essentially the data center of the future.”

(* Disclosure: TheCUBE is a paid media partner for the Dell AI Data Platform Event: Break Through AI With Data event. Neither Dell Technologies Inc., the sponsor of theCUBE’s event coverage, nor other sponsors have editorial control over content on theCUBE or News.)

Image: ChatGPT/ News